African Journal of
Agricultural Research

  • Abbreviation: Afr. J. Agric. Res.
  • Language: English
  • ISSN: 1991-637X
  • DOI: 10.5897/AJAR
  • Start Year: 2006
  • Published Articles: 6007

Full Length Research Paper

Comparison of mapping soybean areas in Brazil through perceptron neural networks and vegetation indices

Carlos Antonio da Silva Junior
  • Carlos Antonio da Silva Junior
  • Geotechnology Applied in Agriculture and Forest (GAAF), State University of Mato Grosso (UNEMAT), Alta Floresta, Mato Grosso, Brazil.
  • Google Scholar
Marcos Rafael Nanni
  • Marcos Rafael Nanni
  • Federal University of Viçosa (UFV), Viçosa, Minas Gerais, Brazil.
  • Google Scholar
Paulo Eduardo Teodoro
  • Paulo Eduardo Teodoro
  • Department of Agronomy (DAG), State University of Maringá (UEM), Maringá, Paraná, Brazil.
  • Google Scholar
Guilherme Fernando Capristo Silva
  • Guilherme Fernando Capristo Silva
  • Federal University of Viçosa (UFV), Viçosa, Minas Gerais, Brazil.
  • Google Scholar
Mendelson Guerreiro de Lima
  • Mendelson Guerreiro de Lima
  • Geotechnology Applied in Agriculture and Forest (GAAF), State University of Mato Grosso (UNEMAT), Alta Floresta, Mato Grosso, Brazil.
  • Google Scholar
Marta Eri
  • Marta Eri
  • University of East Anglia (UEA), Norwich, United Kingdom.
  • Google Scholar


  •  Received: 17 August 2016
  •  Accepted: 11 October 2016
  •  Published: 27 October 2016

Abstract

This study aimed to develop and evaluate the Artificial Neural Networks (ANNs) settings to differentiate and estimate areas of soybean by employing the vegetation index with and without time series. Study area comprises the state of Paraná, South Brazil. The images used to process the ANN were Normalized Difference Vegetation Index (NDVI), Perpendicular Vegetation Index (PVI) and Enhanced Vegetation Index (EVI) indices, including Julian day 017 and Crop Enhanced Index (CEI), which were derived from time series MOD13Q1 product from MODIS sensor Terra satellite. The samples were demarcated into polygon soybean, non-soybean and others (mainly streams). ANN architecture was performed by the module classification employing Multi-layer Perceptron (MLP) artificial neural networks trained by using back propagation algorithm. CEI as a vegetation index with timed series discrimination of soybean areas, pixels with higher than 0.28 rates, proved to be equivalent to ANN to separate soybean areas. Kappa parameter of 0.40 and 0.34 for CEI index and ANN, respectively, it was found in mapped areas. The major and unique contribution of the current study for remote sensing in agriculture was to show that vegetation indices coupled with artificial neural network techniques may improve the results of crop mapping, especially in soybean areas. 
 
Key words: Back propagation neural network, remote sensing, perpendicular vegetation index (PVI), enhanced vegetation index (EVI), crop enhancement index (CEI), time-series.